2631 Background: The number of systemic therapy options for aHCC has increased in recent years. Whilst AB has become the standard of care since its approval in 2020, most patients will have progressed by 12 months and 25% will have no response to AB. In addition, there are no predictive biomarkers to guide aHCC therapy selection. We have previously used phosphoproteomics to identify predictive biomarkers from frozen clinical samples in other cancer types. In this study, we aimed to build a preliminary model of response to AB by performing phosphoproteomic analysis on formalin-fixed and paraffin-embedded (FFPE) resected and Tru-Cut liver biopsies from aHCC patients subsequently treated with AB. Methods: Proteins were extracted from 10x10µm sections of FFPE biopsies obtained at diagnosis from aHCC patients (n=30, aetiology: viral, n=16; non-viral, n=14). Reversal of crosslinks was followed by tryptic digestion and multiple clean-up/enrichment steps. Peptides were quantified by mass spectrometry, and data quality was assessed using multivariate and enrichment analyses. Patients were stratified into ‘good responder’ (GRG, n=20, duration of response (DoR)>7.5 months) and ‘poor responder’ (PRG, n=10, DoR<7.5 month) groups. Features distinguishing the two groups were used to train a random forest response prediction model, which was assessed via cross-validation. Results: To build an AB response prediction model, 40 phosphopeptides were selected based on their ability to distinguish between PRG and GRG patients. These included previously described phosphorylation sites, such as pGSTA1-3S202 and pHSPB1S9, as well as several novel ones. In cross-validation, the model correctly predicted the outcomes of all good (20/20) and of 7/10 poor responders, demonstrating 100% sensitivity, 87% precision and 70% specificity. Overall, our model stratified patients with log-rank p<0.001 and HR<0.1 (Table), with similar performances in both viral (mean DoR 17.1 vs 0 months) and non-viral aetiology (mean DoR 14.4 vs 3.1 months). Interestingly, kinase substrate enrichment analysis revealed significant (p<0.01) modulation of several kinases, such as MAP kinases and PRKCI, between responder groups. Also of note, a subgroup of PRG patients displayed increased activity of the RAF-MEK-ERK pathway, suggesting that these individuals may have shown sensitivity to drugs targeting RAF kinases, such as sorafenib. Conclusions: We have defined a preliminary predictive model of response to AB using phosphoproteomic data from routine FFPE biopsies in aHCC. Following an ongoing validation in independent patient cohorts, this model will address an unmet clinical need for biomarkers of clinical response in aHCC. [Table: see text]